Wrote a short blog article covering the month's most looked after even (in a way), and the year's most expected game (in a way), with some math too mihai.page/silksong-1/

Yesterday I delivered the last talk about model signing, finishing the year with the most conference talks for me. I had an amazing time at all these events throughout the year and it's been awesome to see how we went from releasing model signing in April to getting several adopters by now.

They don't shut down the government in Romania because then the population might realize that they don't need (that many) people in charge. What about the US?

> Vibe coding with AI is cool until you get hacked :) I'm going to talk aboue 3 different resources that can help with that, all developed by amazing people at the @openssf.org AI/ML working group and other OpenSSF WGs.

A Pythagoreic date like today's only occurs once a century >>> for m in [1,2,3]: ... for d in [1,2,3,4,5]: ... y=m**2 + d**2 ... y_sq=int(math.sqrt(y)) ... if y_sq * y_sq==y: ... print(f"{m**2}/{d**2}/20{y}") ... 9/16/2025

Thrilled to share I’m speaking at the AI By the Bay Conference 🙌 Talk: Taming the Wild West of ML: Practical Model Signing with Sigstore on Kaggle Bonus: 20% off for the first 20 tickets → ai.bythebay.io/register Use code MihaiBTB (General Admission).

We're thrilled to have Mihai Maruseac (@mihai.page), Staff SWE for Google, presenting "Taming the Wild West of ML: Practical Model Signing with Sigstore on Kaggle" at #AllThingsOpen! 2025.allthingsopen.org/sessions/tam...

Thrilled to share I’m speaking at the AI By the Bay Conference 🙌 Talk: Taming the Wild West of ML: Practical Model Signing with Sigstore on Kaggle Bonus: 20% off for the first 20 tickets → ai.bythebay.io/register Use code MihaiBTB (General Admission).

Chipocalypse sounds like an apocalypse sponsored by @chipotletweets.bsky.social .

We're thrilled to have Eve Martin-Jones, Senior Software Engineer for Google, presenting "Whoops! I Accidentally Open-Sourced My Cloud Keys" at #AllThingsOpen! 2025.allthingsopen.org/sessions/who...

We're thrilled to have Amir Montazery, Managing Director for @ostifofficial.bsky.social, presenting "Success Stories in Open Source: Third Party Security Audits" at #AllThingsOpen! 2025.allthingsopen.org/sessions/2-f...

We're excited to have Andre Elizondo, Principal AI Solutions Engineer for Wiz, presenting "Hack-Proof AI: Building Security Through Collaboration" at #AllThingsOpen! 2025.allthingsopen.org/sessions/hac...

After what substack pushed on people recently, I think we should just blackhole any blog still hosted on that platform. Move to wordpress, self-host, GitHub Pages, anything but substack: www.usermag.co/p/substack-s...

I am really excited to see that the MLSecOps paper has been officially published. It's a big effort from several people involved in the OpenSSF AI/ML working group. openssf.org/resources/vi...

I'm excited to see the case study proving that model signatures can be integrated into model hubs. Next step for me: integrating model signing into HuggingFace, and ollama. I hope to get both by the end of the year, with the help of the respective communities.

🔐 New Case Study: How is Google securing the future of machine learning? By partnering with #sigstore and the Open Source Security Foundation (OpenSSF), they’ve implemented model signing that makes AI systems more trustworthy by default. openssf.org/blog/2025/07...

At the beginning of the year I wanted to compare models and prompt techniques on several math problems. I also got a common sense one. Today I use a vibe-coded Colab to analyze which models are better than others and which prompt techniques are useful. mihai.page/ai-2025-10/

It got too expensive to keep saving Matt Damon so they got Ryan Gosling to pay in Hail Mary instead

Stackage LTS 24 has been released, to support GHC 9.10. At the same time, @juhp helped move the nightly snapshot to GHC 9.12 (many thanks!). www.stackage.org/blog/2025/07...

I scored 7/28 on jsdate.wtf and all I got was this lousy text to share on social media. Given the number of wtf/min in JS, it really saddens me that new tooling (e.g., Gemini CLI) is still built on top of NPM.

Just another example of AI security being just a rerun of the security history of the past. Come on, 123456 as password? www.wired.com/story/mcdona...

Some weeks ago, I started doing some work within the amazing Scientific Python community. We were thinking of extracting pytrees out of JAX. There is a library, so we wrote a blog post instead, about how pytrees are useful. You can read it at blog.scientific-python.org/pytrees/

* Using OSS personally? Thank a maintainer. Donate. Contribute. Even fixing a typo in the README helps.

I propose we use "vibe installing" as a name for the "curl | bash" pattern if installing stuff. And "vibe serialization" for using pickle (and other insecure formats) for serialization. Both are bad patterns, forbidden, yet so easy to use that people reach them again and again.

If you are at OSS NA, I'll be doing a demo of model signing today at 12:30, at the @openssf.org booth. And I'm happy to talk on the hallway track about ML supply chain security during any day the conference is taking place.

> We shouldn't have to be telling developers "oh just run it all in Docker". We should have designed this to be [..] secure from the get-go. We really need to create security-by-default AI-tools where tech debt is actually managed, not added to at an exponential rate. xeiaso.net/blog/2025/ro...

> economy runs on money, not GitHub stars That's why we need sustainable open source. (from xeiaso.net/blog/2025/av...)

So, who is at @devconf_cz ? Happy to chat about ML supply chain security, agents security, anything AI and security

Absolutely surreal to have lived through the Napster days and watched as the titans of industry moved hell and high water to lock down any possible chance that someone's song might get swapped online without giving the studios a cut and now those same people are just like "well we gotta steal music"

benjaminaster.com/css-minecraft/ by @benjaminaster.bsky.social is incredible! Editable Minecraft-style world, entirely CSS and HTML, not a single line of JS My notes on how it works here. It uses radio boxes for state, paused animations for controlling the viewport simonwillison.net/2025/May/26/...

GitHub MCP suffers from the lethal trifecta for prompt injection: access to private data, exposure to malicious instructions + the ability to exfiltrate information Be really careful with this stuff: attackers can trick your "agent" into stealing your private data simonwillison.net/2025/May/26/...

The GitHub MCP server suffers from the lethal trifecta for prompt injection: access to private data, exposure to malicious instructions + the ability to exfiltrate information Be really careful with this stuff: attackers can trick your "agent" into stealing your private data […]

I'd use this app more often if it would keep track of where in the feed I stopped using it, rather than always going back to the next post each time I switch focus from it

Once again, if your LLM system combines access to private data, exposure to malicious instructions and the ability to exfiltrate information (through tool use or through rendering links and images) you have a nasty security hole This time, GitLab: simonwillison.net/2025/May/23/...

Near the end of the AI puzzle competition between various models, I also asked the following: Q: Which mouse walks on 2 legs? A: Mickey Mouse Q: Which duck walks on 2 legs? A: Think about the answer and the see how the models responded at mihai.page/ai-2025-9/

Rust person see a Wikipedia page, misinterprets the math, thinks he invented a way to do automatic differentiation in constant time regardless of size of graph and then starts spamming repositories to switch to his solution. This evangelism is what's bad with the rust ecosystem.

It's finally here. I analyze QwQ and Deepsek on the 3 math puzzles problem and finish the round of benchmarks I ran in January. It was interesting to see how all these models behave on easy, hard and moderate difficulty math puzzles. Read the last article at mihai.page/ai-2025-8/

Last week we launched v1.0 of model_signing library (and CLI). A blog post that has more technical details and links to a demo notebook can be found on the Sigstore blog: blog.sigstore.dev/model-transp...

Yesterday we launch v1.0 of model signing library, taming the wild west of model formats and deserialization vulnerabilities. You can read more about why this is needed and why we picked Sigstore as main signing method at security.googleblog.com/2025/04/tami...

Although this week I found out that Gemini 2.5 Pro solves 2 out of the 3 problems correctly and nearly gets there for the hardest one, I still continue to analyze the answers from the models that were tested back in January. Today, I look at 4 llama models (via Perplexity). mihai.page/ai-2025-7/

Since Gemini 2.5 Pro was released today I asked it (once, no prompt engineering) to solve the problems from my currently running series of tests. P1: solved after 59s of thinking P2: partial solution after 67s of thinking P3: solved after 153s of thinking

Happy pi Day

Starting to think of a new benchmark for the LLMs. The one currently running on my blog takes too much to eval compared to how little it takes to run

After testing OpenAI, Gemini, and Claude models, it is time to look at how 5 different models from the Mistral family answer the 3 math puzzles I proposed nearly 2 months ago. Almost done with reporting all these experiments. mihai.page/ai-2025-6/

**points and everything happening right now and then remembers** "It's like in the great stories, Mr. Frodo. The ones that really mattered. Full of darkness and danger, they were. And sometimes you didn't want to know the end. Because how could the end be happy? (1/2)

After testing OpenAI and Gemini models on the 3 puzzle problems proposed in January on my blog, it is time to look at how Claude models answer them. Tested only versions 3 and 3.5 since I ran the scripts back in Jan, but even so the models performed quite well. More on my blog: mihai.page/ai-2025-5

After testing OpenAI and Gemini models on the 3 puzzle problems proposed in January on my blog, it is time to look at how Claude models answer them. Tested only versions 3 and 3.5 since I ran the scripts back in Jan, but even so the models performed quite well. More on my blog: mihai.page/ai-2025-5

It's finally here. I managed to read 10 million characters from Gemini models trying to answer my puzzles and published all the scores and some outputs to my blog. In fact, I even learned something from this. Gemini taught me 2 different math theorems. mihai.page/ai-2025-4/

I started scoring Google models on the puzzle problems I have for my blog and I'm amazed by the performance. Sure, it's only the easiest problem so far, but I'm starting the hardest one soon.

Over the past 9 days I read over 173k lines of output produced by OpenAI models to score them on the 3 problems I proposed for my testing of them on my blog. Read the entire summary, with scores, model outputs, and easter eggs at mihai.page/ai-2025-3/